The GPU Was Visible but the NVIDIA Driver Could Not Initialize It
On ZimaOS 1.6.1, a community member installed an NVIDIA RTX PRO 2000 Blackwell and found that nvidia-smi returned “No devices were found.” The PCI device itself was present with ID 10de:2d30, and NVIDIA device nodes had been created. This narrowed the problem to driver initialization rather than a missing PCIe connection or an Ollama container setting.
The diagnostic details also showed NVIDIA kernel and user-space components at version 580.105.08. The open kernel module was loaded, but its log reported an RmInitAdapter failure. NVIDIA's supported chips list for driver 580.105.08 includes device ID 2D30, so the thread did not establish a simple unsupported-device-ID explanation.
The Thread Did Not Confirm a Firmware or Kernel-Parameter Fix
Participants explored firmware paths and NVIDIA module parameters, but later replies corrected several early assumptions. The evidence did not prove that a versioned firmware directory was missing or that a particular module parameter would repair the card. Those ideas should be treated as troubleshooting discussion, not as validated instructions.
The failure also occurred before a Docker workload could use the GPU. Changing an Ollama environment variable or container device mapping would therefore not have fixed the host-level nvidia-smi failure documented on 1.6.1.
ZimaOS 1.6.2 Restored Detection in This Community Case
After updating to ZimaOS 1.6.2, the author reported that the RTX PRO 2000 was recognized and worked with Ollama. This is the strongest resolution recorded in the topic. It is a user-confirmed result for this hardware and installation, not a general compatibility guarantee for every Blackwell GPU or system configuration.
One issue remained: the GPU fan reportedly ran at full speed while idle. The discussion did not reach a confirmed cause or fix for that behavior. GPU recognition and idle fan control should therefore be treated as two separate outcomes.
FAQ
Was Docker or Ollama responsible for the missing GPU?
No evidence in the thread points to Docker as the cause. The host's own nvidia-smi could not initialize the device, so the failure occurred below the container layer.
Did ZimaOS 1.6.2 fix the RTX PRO 2000 problem?
For the topic author, yes: detection returned and Ollama used the GPU after the update. The report does not prove universal support across all systems.
Did the update also fix the fan running at full speed?
No confirmed fan-control solution appears in the discussion. The author still reported full-speed idle fan behavior after GPU recognition was restored.
